CN105067532B - A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab - Google Patents
A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab Download PDFInfo
- Publication number
- CN105067532B CN105067532B CN201510417015.2A CN201510417015A CN105067532B CN 105067532 B CN105067532 B CN 105067532B CN 201510417015 A CN201510417015 A CN 201510417015A CN 105067532 B CN105067532 B CN 105067532B
- Authority
- CN
- China
- Prior art keywords
- wave band
- scab
- pixel
- high spectrum
- spectrum image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab, this method includes：(1) high spectrum image of rape leaf to be measured is taken, then selects the pixel of the scab part of the high spectrum image；(2) it calculates and obtains wave band of the pixel in 681.95nm~748.23nm wave bands than operation values and wave band calculus of differences value；(3) housebroken least square method supporting vector machine model will be inputted than the feature vector that operation values and wave band calculus of differences value are formed by spectral value, the wave band of the pixel at 555.29nm, scab type is judged according to output result.The present invention analyzes from the high spectrum image information with sclerotiniose, gray mold early stage scab and obtains feature difference wave band, by characteristic wavelength, wave band than operation values and wave band calculus of differences value and least square method supporting vector machine combination, the Accurate classification of sclerotiniose and gray mold early stage scab is realized, differentiates precision higher.
Description
Technical field
The present invention relates to Digital Agriculture technical field more particularly to a kind of discriminating sclerotinia sclerotiorums and gray mold early stage scab
Method.
Background technology
Rape is the important oil crops in China, and the stress of various diseases is often subject in planting process, leads to the underproduction.
The Major Diseases of rape have sclerotiniose and gray mold etc..Sclerotiniose caused by sclerotinite is infected, by the pathogen of Botrytis cinerea invaded by gray mold
Caused by dye.During two kinds of disease infestation rape leaf initial stages, water stain shape scab is all shown on rape leaf, it is difficult to distinguish two kinds of diseases
Evil be easy to cause erroneous judgement and misses the best prevention time.If accurately identifying both diseases with a kind of method of quick nondestructive,
Peasant household can then be instructed to apply the pesticide for specific diseases, improve efficiency and to greatest extent environmental protection.
Image processing techniques is combined by hyper-spectral image technique with spectral analysis technique, becomes Digital Agriculture neck in recent years
The hot spot of domain research.It has adhered to the characteristics of non-destructive testing of spectrum and hyperspectral technique, while utilizes image processing techniques energy
Enough early diseases for more accurately identifying plant.Hyper-spectral image technique has been used for fruit quality discriminating, plant nutrition ingredient
Discriminating, the discriminating of plant diseases degree and discriminating of infected plant and healthy plant etc..
Moshou using visible and near infrared band hyperspectral information differentiate wheat healthy leaves, infection yellow rust blade and
Nitrogen stress blade, with quadratic discriminatory analysis (quadratic discriminant analysis, QDA) and self-organizing map neural
Two kinds of sorting techniques of network (self-organizing map neural net) differentiate this three classes blade, deficiency be by
There are unstability (Moshou D., Bravo C., Wahlen S., et for the canopy reflectance spectra of disease infection
al.Simultaneous identification of plant stresses and diseases in arable crops
based on a proximal sensing system and Self-Organising Neural Networks：
Proceedings of 4th European Conference on Precision Agriculture, 2003 [C]
.Berlin：Stafford J., Werner A., 2003：425-432.Moshou D., Bravo C., Wahlen S., et
al.Simultaneous identification of plant stresses and diseases in arable crops
Using proximal optical sensing [J] .Precision Agriculture, 2006,7 (3)：149-164.).
The canker of grape fruit fruit is separated (Qin Jianwei, Burks by Qin using hyperspectral information from healthy fruit
T.F., Kim M.S., et al.Citrus canker detection using hyperspectral reflectance
imaging and PCA based image classification method[J].Sensing Instrument for
Food Quality and Safety, 2008,2 (3)：168-177.), Qin further by the canker of grape fruit fruit from its
It is separated in the disease fruit of his type, the sorting technique of the research has used 99 wave bands to participate in analysis, and nicety of grading is higher, still
Relatively time consuming (Qin Jianwei, Burks T.F., Ritenour M.A., the et al.Detection of citrus of operation
canker using hyperspectral reflectance imaging with spectral information
Divergence [J] .Journal of Food Engineering, 2009,93 (2)：183-191.).Li Jiangbo principal component
Analysis is combined navel orange ulcer fruit than algorithm with wave band and is separated from the other kinds of disease fruit of navel orange, relative to Qin's
Research reduces operand, and (Li Jiangbo, Rao Xiuqin answer justice is refined is waited to detect navel orange ulcer [J] agricultures based on high light spectrum image-forming technology
Industrial engineering (IE) journal, 2010,26 (8)：222-228.).
But differentiate the research of the extremely similar different diseases of the symptom of same plant using hyper-spectral image technique
It is less.T.Rumpf hyperspectral information combination vegetation index (Vegetation Indices) and support vector machines (Support
Vector Machines) by the three classes beet disease such as beet healthy leaves and rubiginose, Cercospora leaf spot and powdery mildew
Blade is distinguished, total to differentiate that precision is more than 86%, but precision is undesirable when three kinds of diseases are differentiated two-by-two (Rumpf T.,
Mahlein A.K., Steiner U., et al.Early detection and classification of plant
diseases with Support Vector Machines based on hyperspectral reflectance[J]
.Computers and Electronics in Agriculture, 2010,74 (1)：91-99.).This is high spectrum image skill
Art not yet solves the problems, such as very well for one in differential plant disease research.
Invention content
The present invention provides a kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab, this method can accurately reflect
The early stage scab of sclerotiniose and gray mold on other rape leaf differentiates that precision is high.
A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab, including：
(1) high spectrum image of rape leaf to be measured is taken, then selects the pixel of the scab part of the high spectrum image；
(2) wave band for obtaining the pixel in 681.95nm~748.23nm wave bands is calculated to transport than operation values and wave band difference
Calculation value；
(3) by by the pixel in the spectral value of 555.29nm, the wave band than operation values and wave band calculus of differences value
The feature vector of composition inputs housebroken least square method supporting vector machine model, judges scab type according to output result.
Specifically, the training method of the least square method supporting vector machine model, includes the following steps：
(1) take infection sclerotiniose, gray mold rape leave high spectrum image scab part pixel, composing training collection；
(2) pixel in training set is calculated to transport than operation values and wave band difference in the wave band of 681.95nm~748.23nm wave bands
Calculation value；
(3) with by spectral value of the pixel in training set in 555.29nm, the wave in 681.95nm~748.23nm wave bands
Section is used as input than the feature vector that operation values and wave band calculus of differences value are formed, using the type of scab where pixel as output,
Training least square method supporting vector machine model.
It is described take infection sclerotiniose, gray mold rape leave high spectrum image scab part pixel method, including with
Lower step：
(1) mycelia block is inoculated on the rape leaf of health；
(2) high spectrum image of infected rape leaf is obtained；
(3) region in high spectrum image where mycelia block is removed；
(4) region where healthy mesophyll in high spectrum image is removed, remainder is the scab office of high spectrum image
Portion.
Preferably, the method in the region in the removal high spectrum image where mycelia block is：By high spectrum image
It is poor that spectral value and 679.42nm spectral value of each pixel at 549.10nm are made, and mycelia in high spectrum image is distinguished according to difference
Region where block.
Preferably, healthy mesophyll pixel and the scab office in high spectrum image are distinguished according to the spectral value at 512.05nm
Portion's pixel.
Preferably, the hyper parameter γ of the least square method supporting vector machine model is 0.79214, kernel functional parameter δ2For
4.4687。
The present invention also provides a kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab, including：
(1) high spectrum image of rape leaf to be measured is taken, then selects the pixel of the scab part of the high spectrum image；
(2) wave band for obtaining the pixel in 681.95nm~748.23nm wave bands is calculated to transport than operation values or wave band difference
Calculation value；
(3) it determines the threshold value of classification, scab type is determined than operation values or wave band calculus of differences value according to the wave band.
Specifically, the threshold value determination method is：The wave band of known sample pixel is calculated than operation values or wave band difference
Operation values draw receiver operation indicatrix, select best threshold value.
Preferably, when being classified according to wave band than operation values, threshold value 4.5；It is carried out according to wave band calculus of differences value
During classification, threshold value 0.6.
Compared with prior art, the invention has the advantages that：
The present invention analyzes from the high spectrum image information with sclerotiniose or gray mold early stage scab and obtains feature difference
Wave band by characteristic wavelength, wave band than the combination of operation values and wave band calculus of differences value and least square method supporting vector machine, is realized
The Accurate classification of sclerotiniose and gray mold early stage scab compared with conventional principal component analytical method, differentiates precision higher.
Description of the drawings
Fig. 1 is the structure diagram of high spectrum image system of the present invention；
Wherein, 1. hyperspectral imager；2. line source；3. high spectrum image acquires and analysis software；4. support board；5. electricity
Control displacement platform.
Fig. 2 is the high spectrum image of sclerotinia sclerotiorum spot and grey mold scab；
I is the rape leaf for being inoculated with sclerotium scab；II is the rape leaf for being inoculated with grey mold scab；III is the partial enlargement of I
Figure；IV is the partial enlarged view of II.
Fig. 3 is the average light of sclerotinia sclerotiorum and the 1st day (A), the 2nd day (B) and the 3rd day (C) scab after gray mold morbidity
Spectrogram.
Fig. 4 is mycelia block and the averaged spectrum curve of mesophyll tissue；
A is the curve of spectrum of mycelia block and mesophyll；B is that the spectrum of healthy mesophyll and infected mesophyll (two kinds of scabs) is bent
Line.
Fig. 5 is the averaged spectrum curve of sclerotinia sclerotiorum spot, grey mold scab and the two difference and healthy mesophyll.
Fig. 6 is the band math result of sclerotinia sclerotiorum and gray mold in the method for the present invention；
Wherein, upper row's picture is sclerotium scab, and lower row's picture is grey mold scab；
A1, A2 are true color picture；B1, B2 are mask picture；C1, C2 are wave band than operation picture；D1, D2 and E1, E2
For wave band calculus of differences picture.
Fig. 7 is classification results of the wave band than operation image；
Wherein, the wave band that left side is the 1st day, the 2nd day and the 3rd day after falling ill is than the ROC curve of operation image；Right side is hair
The wave band of the 1st day after being ill, the 2nd day and the 3rd day is than the nicety of grading curve of operation image.
Fig. 8 is the classification results of wave band calculus of differences image；
Wherein, left side is the ROC curve of the wave band calculus of differences image of the 1st day, the 2nd day and the 3rd day D after falling ill；Right side
Nicety of grading curve for the wave band calculus of differences image of the 1st day, the 2nd day and the 3rd day after morbidity.
Fig. 9 is the scatter plot of preceding 3 principal components of sclerotinia sclerotiorum and gray mold scab high spectrum image.
Specific embodiment
The present invention is further elaborated with attached drawing With reference to embodiment.
Preparation before experiment
1st, the preparation of experiment material
In order to simulate natural occurrence state, this experiment is blade inoculation germ with mycelia block small as possible.Use culture medium
(Potato dextrose agar medium, PDA) cultivates sclerotiniose mycelia and gray mold mycelia, and the circle of 5 millimeters of diameter is made
Shape mycelia block.Two kinds of diseases are inoculated on the rape leaf of same growth period with mycelia block, are chosen on each blade
Analogous location sticks a little for mycelia block.
60 blade samples of this Preparatory work of experiment, wherein 30 for being inoculated with sclerotiniose, 30 for being inoculated with gray mold,
Using Non in vitro vaccination ways.Manually incubator keeps 25 DEG C of rapeseed plants constant temperature, constant humidity 85%, the mycelia after 24 hours
Start to show symptom around block, the area of scab is all very small in 3 days after morbidity and two kinds of scabs are closely similar, i.e.,：Hair
Sick initial stage is all shown as water stain shape scab, it becomes difficult to distinguish.
2nd, the composition of high spectrum image system
This experiment Hyperspectral imager used mainly by spectrometer (Imspector V10E, SPECIM, Finland,
Sampling interval 1nm, spectral resolution 6nm), a frame pixel is 1392 × 1040 CCD camera (Hamamatsu, Japan),
Optical filter (OBF570, SPECIM, Finland), camera lens (SCHNEIDER, Germany), a set of complete two branch's light of spectrum of 150W
Fine line source (Illumination, USA), one group of automatically controlled displacement platform and mating image capture software are formed, as shown in Figure 1.It adopts
The high-spectrum of collection seem a size be 672 × 409 (space dimension) × 512 (spectrum dimension) three-dimensional data structure, spectrum model
It encloses for 380~1030nm, totally 512 wave bands.
3rd, high spectrum image calibration method
Before sample image is obtained, one group of white standard picture and black standard picture are first acquired, white standard picture is by sweeping
It retouches the standard white rectification plate that reflectivity is 99% to obtain, black standard picture is obtained by covering the scanning of image acquisition device lens cap.
Calibration image is calculated by formula (1) in collected all original high spectrum images.
R=(I-Idark)/(Iwhite-Idark) (1)
Wherein, R is the calibration image of sample, and I is the original image of sample, IdarkIt is black standard picture, IwhiteIt is white mark
Quasi- image.
4th, Spectral Characteristics Analysis
Rape high spectrum image is obtained by above-mentioned high spectrum image acquisition system, as shown in Fig. 2, Fig. 2 is to take picture number
According to the scab position true color high spectrum image that middle 640nm, 550nm and 460nm are formed, wherein, Fig. 2 (I) is inoculation sclerotiniose
The rape leaf of spot, Fig. 2 (II) are the rape leaf for being inoculated with grey mold scab.
In order to which the method for the present invention is enable preferably to be promoted in production practices, overlook and adopt in each panel height spectrum picture
Collect a complete rapeseed plants, then analyzed for the scab position on blade.
The scab very little caught an illness on initial stage blade is needed image magnification clearly to show scab, Fig. 2 (III) and Fig. 2
(IV) be respectively Fig. 2 (I) and Fig. 2 (II) partial enlarged view, the A points in Fig. 2 (III) and Fig. 2 (IV) is in scab regions
One point, B points are a point in mycelia block region, and the C points in Fig. 2 (III) are the mycelia area that mycelia block edge grows out
One point in domain.
Since temperature and humidity conditions are suitable, the mycelia growth in Fig. 2 is very fast.In Fig. 2, the mycelia block shape of sclerotiniose and gray mold
Shape size is identical, and the time cultivated in culture dish is identical, and hyphae length is similar, the spot pattern difference of formation may be because
The tightness degree that the bumps small for blade face and the natural form of blade make mycelia block be bonded with blade face has nuance, causes bacterium
There is difference in sprawling direction of the silk on blade, has no effect on the progress of experiment.
Sclerotinia sclerotiorum and gray mold scab do not have significant difference after morbidity in 3 days, study in this period
Scab sorting technique is all meaningful.So after this experiment analyzes hyper-spectral image technique to both disease morbidities respectively
The distinguishing ability of the 1st day, the 2nd day and the 3rd day.The averaged spectrum of scab pixel is as schemed in sclerotinia sclerotiorum and gray mold 1~3 day
Shown in 3.
5th, the extraction of area-of-interest
The collected sample hyperspectral information of present system is in the range of 380~1030nm of wavelength, due to close to two
Data in the wave-length coverage of end have larger noise, influence later data processing, so only extracting 359 of 400~850nm of wavelength
The hyperspectral information of wave band is analyzed.
Region of interest (Regions of interest, ROI) is for by interested research object pixel and other objects
Pixel distinguishes, and makes further work only to interested regional implementation.
The area-of-interest of this experiment is the infected position of mesophyll, so needing region of interest from healthy mesophyll and bacterium
Silk block is separated in region.
Due to mycelia block pixel the curve of spectrum and health or infected mesophyll pixel the curve of spectrum in shape all
There are notable difference, mesophyll pixel spectra curve is in monotone decreasing in the range of 549.10nm to 679.42nm, and mycelia block picture
The plain curve of spectrum is in monotonic increase in the range, as shown in Figure 4 A.So this experiment is first subtracted with 549.10nm wave band datas
679.42nm wave band datas so that notable difference is presented in mesophyll and mycelia block pixel, so that mycelia block is distinguished.
Next healthy mesophyll and the mesophyll pixel by two kinds of disease infections are distinguished.From infected mesophyll and healthy mesophyll
In the curve of spectrum of pixel observation learn, in two wavelength bands of 450nm to 525nm and 580nm to 700nm, health and
Infected mesophyll spectroscopic data is not overlapped (such as Fig. 4 B), it is possible to for distinguishing healthy mesophyll and by two kinds of disease infections
Mesophyll pixel, in order to make the difference of the spectral value of two class pixels big as possible, this experiment selects 512.05nm wave bands to distinguish two class leaves
Meat pixel (threshold value=0.16).
1 band math of embodiment and characteristic wave bands-least square method supporting vector machine (LS-SVM)
By observing sclerotiniose and gray mold spectral differences (such as Fig. 5), sclerotiniose and ash in 675-750nm regions are found
There are larger differences for the slope of the mildew curve of spectrum, can distinguish them using this difference.Sclerotiniose and gray mold simultaneously
The maximum value of spectral differences is at 748.23nm, and minimum value is at 681.95nm, therefore, wave is done to the two wave bands
Section can effectively distinguish two kinds of scabs than operation and wave band calculus of differences.
It chooses the two wave bands and carries out operation, formula is as follows：
Band ratio=R748.23/R681.95 (2)
Difference=R748.23-R681.95 (3)
In formula, band ratio be image band than operation as a result, difference be image difference operation as a result,
R748.23 is the spectral reflectance value of 748.23nm single band image pixels, and R681.95 is 681.95nm single band image pixels
Spectral reflectance value.
In plant spectral, the SPECTRAL REGION of green wavelength 550nm or so can effectively describe the feature of plant leaf blade.Such as
Shown in Fig. 5, the curve of spectrum of two kinds of scabs of rape leaf forms a reflection peak, and two kinds of scab spectral differences at 555.29nm
The wave crest of value curve is also appeared at the wave band, so this wave band can be used as characteristic wave bands to differentiate two kinds of scabs.
To further improve the precision of discriminating, herein with the wave band of two kinds of scabs than operation values, wave band calculus of differences value and
Spectral reflectance value at 555.29nm in feature vector input module as carrying out classification model construction, it is desirable to access higher point
Class precision.
In order to be comparable the discrimination method in the present embodiment, modeling algorithm uses least square method supporting vector machine
(LS-SVM), kernel function uses radial basis function (RBF Kernal), and γ and δ are obtained by cross verification2Optimal combination：
γ=0.79214, δ2=4.4687.The discriminating precision of rape leaf sclerotiniose and gray mold the 1st, 2,3 day scab of morbidity, such as table
Shown in 1.
Rape scab of the table 1 based on band ratio, wave band difference value and 555.29nm reflected values differentiates precision
2 band maths of embodiment-receiver operation indicatrix (Band Math-ROC)
For the ease of postorder image analysis, the result of region of interesting extraction is made into mask data, region of interest is set
Domain is 1, and other regions are 0.
The present embodiment proposes band math-receiver operation indicatrix (Band Math-ROC) algorithm to differentiate two kinds
Scab, the band math are divided into wave band than algorithm and wave band difference algorithm, it is only necessary to which 2 wave bands participate in operation, can drop significantly
Low calculation amount.
It is analyzed after applying mechanically mask to wave band ratio and wave band calculus of differences result, as shown in Figure 6.The present embodiment trial pair
Wave band distinguishes two kinds of scabs than the result with wave band calculus of differences into row threshold division.The selection of threshold value is an important step
Suddenly, the threshold value of selection is close to the first kind, then the probability that the second class can be correctly validated is just high, while the first kind can correctly be known
Other probability can be low；If threshold value, close to the second class, result is opposite.So can should correctly it be classified ensureing two classes as possible
In the case of choose one compromise value.
Receiver operation indicatrix (Receiver operating characteristic curve, ROC) can have
Effect describes the class separating capacity of the arbitrary criterion in entire range of choice.Each point reflects identical sensitivity on curve,
They are all the reactions to same signal stimulus under several different criterions；With false positive rate (false probability P (y/N))
For abscissa, with true positive rate (hit probability P (y/SN)) for ordinate, according to sample under the conditions of particular stimulation due to using
The curve that the Different Results that different criterions obtains are drawn.If two class sample distributions are completely superposed, ROC curve and y
=x straight lines overlap；If two class sample distributions are kept completely separate, ROC curve is overlapped with y=1 straight lines；Therefore in entire criterion
In the range of, it is moved in the triangle that ROC curve is all surrounded in above-mentioned two straight lines with y-axis.
Region (area under curve, AUC) below receiver operation indicatrix can having as classification performance
Index is imitated, the more big then classifier performance of AUC value is better.It is put on usual ROC curve to (0,1) apart from the corresponding judgement of shortest point
The classification performance of standard is best.
So the present embodiment selection receiver operation feature (receiver operating characteristic,
ROC) curve carrys out preferred classification thresholds.Formula (4) and formula (5) are the calculation formula of true positive rate and false positive rate respectively, formula
Middle TP represents true positives, and FN represents false negative, and FP represents false positive, and TN represents true negative.
TPR=TP/ (TP+FN) (4)
FPR=FP/ (FP+TN) (5)
In the present embodiment, if sclerotium scab sample is the positive, grey mold scab sample is feminine gender, then true positive rate is sclerotiniose
The correct recognition rata of spot, false positive rate are the error recognition rate of sclerotium scab.
In experimentation, divide the sample of sclerotiniose and the morbidity the 1st, 2,3 day of grey mold scab respectively with a series of threshold value
Image often takes a threshold value just to form the data point on a ROC curve.The standard that threshold value is chosen is can to get over true positive rate
Close to 1, while false positive rate is more outstanding closer to 0 threshold value.
So the data point near coordinate (0,1) point on ROC curve corresponds to optimal threshold.Wave band is than operation image
ROC curve and nicety of grading curve as shown in fig. 7, the ROC curve and nicety of grading curve graph of wave band calculus of differences image such as
Shown in Fig. 8, optimal threshold is corresponded to the data point of dashed lines labeled in figure.
From Fig. 7 and Fig. 8 it is obvious that wave band than operation optimal threshold for 4.5, wave band calculus of differences it is best
Threshold value is 0.6.In the case where setting optimal threshold, sclerotinia sclerotiorum spot and the discriminating precision of grey mold scab are as shown in table 2.
Table 2 differentiates precision based on wave band than the rape scab of operation values and wave band calculus of differences value
Comparative example 1
Principal component analysis (Principal component analysis, PCA) is on the basis that high spectrum image is calibrated
On to the effective ways of image dimensionality reduction.It can be made up of the recombinant of each spectral band data dimension is lower, feature more
Apparent spectrum picture.Preceding n (n ＜ ＜ all-waves hop count) a compositional data after recombinant can almost express original high-spectrum
Information as in more than 90% reduces calculation amount and the time of postorder analysis.But it is directly distinguished with preceding n compositional data
Go out two kinds of scabs also to have difficulties, it may be possible to because the characteristics of image of principal component analysis concern is not the feature for testing concern.For
The one of characteristics of objects that can most express in preceding n compositional data and be paid close attention in experiment has been found in further separation scab, this experiment
A compositional data.
Specific analytical method is as follows：
Sclerotiniose and gray mold scab pixel principal component are analyzed in this experiment, and preceding 3 principal components (are respectively PC1, PC2
And PC3) contain the information that high spectrum image is more than 90%.Scab pixel is individually placed to preceding 3 principal components to form two-by-two
Coordinate system in, as shown in Figure 9.It will be apparent that two kinds of scabs can not be detached at all along the direction of PC1 axis, PC2 separation scabs
Ability is higher than PC3, is further analyzed so choosing PC2 data.
In principal component analysis, each principal component is all original single band images by participation operation by linear
It is composed, weight coefficient w is bigger in this linear combination, and corresponding single band image gets over the contribution of the principal component
Greatly, i.e.,：The size of the weight coefficient w of each wave band represents the size that the wave band contributes compositional data, so for quilt
The compositional data of choosing, several larger original wave bands of weight coefficient are capable of the feature of effective expression experiment concern.
In order to find out the single band image maximum to PC2 contributions, weight of this experiment by PC2 on original single band image
Coefficient is ranked up by size.
Shown in weight coefficient such as formula (6)：
Wherein, PC represents the compositional data after recombinant, and B represents to participate in the original wave band of operation, and m is the number of original wave band
Amount, w represent weight coefficient.
It sorts according to weight coefficient, takes characteristic wavelength of 5 wave bands of weight coefficient maximum as two kinds of scabs of separation, it
Be 556.54nm, 670.55nm, 673.08nm, 485.03nm, 487.48nm respectively.These characteristic wavelengths can be used as classification
The input variable of model.
Scab disaggregated model is established using preferred 5 characteristic wave bands as the input variable of LS-SVM.Sample is collected to modeling
This is the 1st day after morbidity, the 2nd day, the 3rd day high-spectral data obtained establish LS-SVM disaggregated models, kernel function is radial direction base
Function (RBF Kernal), γ and δ are obtained by cross verification2Optimal combination：γ=0.79214, δ2=4.4687.
Classified with the model to verification collection sample, precision is as shown in table 3.
The rape scab of 3 feature based wavelength of table and LS-SVM differentiate precision
Embodiment 1,2 and comparative example 1 the result shows that, embodiment 1, embodiment 2 rape scab differentiate that precision is above
Comparative example 1, reason may is that using the method that characteristic wavelength is found in principal component analysis be for whole blade, including scab and
Healthy pixel, it is impossible to the effectively feature of extraction scab.
In addition, the discriminating precision of wave band calculus of differences, again higher than wave band than operation, main cause is that wave band is reduced than operation
The distance of Different categories of samples, and calculus of differences remains the metric space of all kinds of；555.29nm wave band is to discriminate between rape leaf
The validity feature of two kinds of diseases, which is combined with band ratio, wave band difference value can be more effectively by two
Kind of disease sample classification, the results showed that the rape scab based on these three values differentiate model accuracy really higher than wave band than operation and
The scab nicety of grading of wave band calculus of differences.
Result of study explanation, can using high spectrum image information combination band math and least square method supporting vector machine model
Effectively to differentiate sclerotinia sclerotiorum and gray mold early stage scab.
Claims (1)
1. a kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab, including：
(1) high spectrum image of rape leaf to be measured is taken, then selects the pixel of the scab part of the high spectrum image；
(2) it calculates and obtains wave band of the pixel in 681.95nm~748.23nm wave bands than operation values and wave band calculus of differences
Value；
(3) it will be made of spectral value, the wave band of the pixel at 555.29nm than operation values and wave band calculus of differences value
Feature vector input housebroken least square method supporting vector machine model, scab type is judged according to output result；
The training method of the least square method supporting vector machine model, includes the following steps：
(A) take infection sclerotiniose, gray mold rape leave high spectrum image scab part pixel, composing training collection；
It is described take infection sclerotiniose, gray mold rape leave high spectrum image scab part pixel method, including following step
Suddenly：
(a) mycelia block is inoculated on the rape leaf of health；
(b) high spectrum image of infected rape leaf is obtained；
(c) region in high spectrum image where mycelia block is removed；Specific method is：Each pixel of high spectrum image is existed
It is poor that spectral value and 679.42nm spectral values at 549.10nm are made, the area according to where difference distinguishes mycelia block in high spectrum image
Domain；
(d) the healthy mesophyll pixel and scab local pixel in high spectrum image are distinguished according to the spectral value at 512.05nm；It goes
Except the region where mesophyll healthy in high spectrum image, remainder is the scab part of high spectrum image；
(B) calculate training set in pixel 681.95nm~748.23nm wave bands wave band than operation values and wave band calculus of differences
Value；
(C) with by spectral value of the pixel in training set in 555.29nm, the wave band ratio in 681.95nm~748.23nm wave bands
The feature vector that operation values and wave band calculus of differences value are formed is as input, using the type of scab where pixel as output, training
Least square method supporting vector machine model；The hyper parameter γ of the least square method supporting vector machine model is 0.79214, kernel function ginseng
Number δ2It is 4.4687.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510417015.2A CN105067532B (en) | 2015-07-15 | 2015-07-15 | A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510417015.2A CN105067532B (en) | 2015-07-15 | 2015-07-15 | A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105067532A CN105067532A (en) | 2015-11-18 |
| CN105067532B true CN105067532B (en) | 2018-07-06 |
Family
ID=54496956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510417015.2A Active CN105067532B (en) | 2015-07-15 | 2015-07-15 | A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105067532B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105784604A (en) * | 2016-03-04 | 2016-07-20 | 北京大学深圳研究生院 | Plant declination level detecting method |
| CN105931223B (en) * | 2016-04-13 | 2019-05-21 | 浙江大学 | The dividing method in maize portion in high spectrum reflection image based on wave band ratio method |
| CN105975966A (en) * | 2016-04-21 | 2016-09-28 | 南京农业大学 | Rice grain mildew nondestructive test method |
| CN106777845A (en) * | 2017-03-22 | 2017-05-31 | 南京农业大学 | The method that sensitive parameter builds wheat leaf blade powdery mildew early monitoring model is extracted based on subwindow rearrangement method |
| CN110501346B (en) * | 2019-09-24 | 2021-02-19 | 江南大学 | Quality detection system based on multispectral image acquisition technology |
| CN110646359A (en) * | 2019-10-17 | 2020-01-03 | 黑龙江八一农垦大学 | Small red bean rust detection method and system |
| CN113237864A (en) * | 2021-03-23 | 2021-08-10 | 浙江大学山东工业技术研究院 | Method for rapidly diagnosing dynamic development of sclerotinia rot of rape leaves based on spectral imaging technology |
| CN113237863A (en) * | 2021-03-23 | 2021-08-10 | 浙江大学山东工业技术研究院 | Method for rapidly diagnosing sclerotinia sclerotiorum infection of rape based on spectrum |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102008058422A1 (en) * | 2008-11-21 | 2010-05-27 | Precitec Itm Gmbh | Monitoring laser machining process to be carried out on workpiece, comprises detecting actual measuring values by sensor, which monitors the laser machining process, and determining actual characteristic values from actual measuring values |
| CN103018180A (en) * | 2012-12-11 | 2013-04-03 | 江苏大学 | Method and device for detecting cotton diseases based on multisource light information technology |
-
2015
- 2015-07-15 CN CN201510417015.2A patent/CN105067532B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102008058422A1 (en) * | 2008-11-21 | 2010-05-27 | Precitec Itm Gmbh | Monitoring laser machining process to be carried out on workpiece, comprises detecting actual measuring values by sensor, which monitors the laser machining process, and determining actual characteristic values from actual measuring values |
| CN103018180A (en) * | 2012-12-11 | 2013-04-03 | 江苏大学 | Method and device for detecting cotton diseases based on multisource light information technology |
Non-Patent Citations (5)
| Title |
|---|
| Detection of citrus canker using hyperspectral reflectance imaging with;jianwei Qin，et al;《Journal of Food Engineering》;20091231(第93期);183-191 * |
| The use of the area under the roc curve in the evaluation of machine learning algorithms;ANDREW P.BRADLEY;《pattern recognition》;19971231;第30卷(第7期);1145-1149 * |
| 基于光谱和多源波谱成像技术的植物灰霉病快速识别的方法研究;陈双双;《中国优秀硕士学位论文全文数据库 农业科技辑》;20120715(第7期);49-54 * |
| 基于高光谱和图像处理技术的油菜病虫害早期监测方法和机理研究;赵芸;《中国博士学位论文全文数据库 信息科技辑》;20130615(第6期);53-55 * |
| 郭俊先.基于高光谱成像技术的棉花杂质检测方法的研究.《中国博士学位论文全文数据库 信息科技辑》.2011,(第7期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105067532A (en) | 2015-11-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105067532B (en) | A kind of method for differentiating sclerotinia sclerotiorum and gray mold early stage scab | |
| CN110132862B (en) | Construction method and application of disease index special for wheat scab detection | |
| Fuentes et al. | Automated grapevine cultivar classification based on machine learning using leaf morpho-colorimetry, fractal dimension and near-infrared spectroscopy parameters | |
| CN101881726B (en) | Nondestructive detection method for comprehensive character living bodies of plant seedlings | |
| CN101692037B (en) | Method for analyzing chlorophyll distribution on surface of leaves of plant by hyperspectral image and independent component | |
| CN110082298B (en) | Hyperspectral image-based wheat variety gibberellic disease comprehensive resistance identification method | |
| CN105021529A (en) | Spectrum information and image information fusing crop plant disease and insect pest identifying and distinguishing method | |
| CN104697943A (en) | Rape water stress nondestructive testing method | |
| CN112287886B (en) | Wheat plant nitrogen content estimation method based on hyperspectral image fusion map features | |
| CN110363125A (en) | Using the method for Model Transfer identification different cultivars Citrus Huanglongbing pathogen | |
| Mengistu et al. | An automatic coffee plant diseases identification using hybrid approaches of image processing and decision tree | |
| CN111832507A (en) | Wheat head top spectrum information-based wheat scab remote sensing identification method | |
| CN103528967A (en) | Hyperspectral image based overripe Lonicera edulis fruit identification method | |
| Arakeri et al. | Analysis of late blight disease in tomato leaf using image processing techniques | |
| CN113570538A (en) | Blade RGB image skewed distribution parameter information acquisition and analysis method | |
| Gaikwad et al. | Multi-spectral imaging for fruits and vegetables | |
| Zhang et al. | Chlorophyll content detection of field maize using RGB-NIR camera | |
| Suzuki et al. | Image segmentation between crop and weed using hyperspectral imaging for weed detection in soybean field | |
| CN102521830B (en) | Optimum band selection method for hyperspectral images of canopy of crop under disease stress | |
| Du et al. | Diagnostic model for wheat leaf conditions using image features and a support vector machine | |
| Husin et al. | Distribution map of oil palm fresh fruit bunch using LiDAR | |
| De Silva et al. | Plant disease detection using deep learning on natural environment images | |
| CN116824171A (en) | Method for selecting hyperspectral tongue picture image wave band of traditional Chinese medicine and related device | |
| Wang et al. | Characterization and classification of urban weed species in northeast China using terrestrial hyperspectral images | |
| Zhou et al. | Hyperspectral imaging technology for detection of moisture content of tomato leaves |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |